Perfect Cuboids and the Box Variety Ernst Kani Queen s University Québec-Maine Conference, Laval University 27 Sept 2014
Outline 1. Introduction 2. Early History 3. New Ideas (using Arithmetic Geometry) 4. The Bombieri-Lang Conjecture 5. Further Results 6. Diagonal Quotient Surfaces 7. Modular Correspondences 8. Mazur s Question
1. Introduction Consider a rectangular box (also known as cuboid): y w z a x b c By Pythagoras we have the following relations: (1) a 2 + b 2 = x 2 (2) b 2 + c 2 = y 2 (3) a 2 + c 2 = z 2 (4) a 2 + b 2 + c 2 = w 2
1. Introduction 2 A rational cuboid is a solution of (1) (3) with a, b, c, x, y, z Q +. A perfect cuboid is a solution of (1) (4) with a,..., w Q +. Problem 1: (Sanderson, 1740, Euler 1770) Find (parametric families of) rational cuboids. Open Problem 2: Are there any perfect cuboids? Open Problem 3: Are there at most finitely many perfect cuboids?
2. Early History P. Halcke, 1719: observed that (a, b, c) = (44, 240, 117) defines a rational cuboid. (This is the smallest one!) Sanderson, 1740, Euler 1770: Using pythagorean triples, Sanderson and Euler gave in their respective texts Elements of Algebra a systematic method for constructing rational cuboids. These are now called Euler cuboids; cf. Euler, Elements of Algebra, II, Art. 238 (p. 443). A. Martin, 1894: In L intermediaire des mathématiciens vol. 1 (1894), p. 214, Artemas Martin (Washington) posed Problem 2 as Question 361. A solution was offered by Brocard (1895), which was criticized by Tannery (1896), as is mentioned in L. Dickson s History of Number Theory, vol. II, ch. XVII.
2. Early History - 2 H. Olsen, 1916: Problem 254 of the American Mathematical Monthly 23 (1916) (proposed by H. Olsen, Chicago) requires you to find all perfect cuboids. V. Spunar, 1917 submits a solution (which is published) that no such cuboids exist. (This solution is mentioned but is not criticized by Dickson, although it should be.) M. Kraitchik, 1954: establishes some congruence conditions for the sides a, b, c of a perfect cuboid. J. Leech, 1977: studies cuboids for which a, b, c and only 3 of x, y, z, w are rational (suggested by M. Gardner, 1970). I. Korec, 1992: Using his earlier results and a computer search, Korec shows that for a perfect cuboid max(a, b, c) > 4 10 9.
3. New Ideas (using Arithmetic Geometry) Question: Is there a geometric reason why it is easy to find lots of rational cuboids but difficult to find perfect ones? A. Bremner, 1988: studies the projective surface V P 5 defined by equations (1) (3). Thus, the rational points V (Q) of V (with abc 0) correspond to rational cuboids. He constructs another surface W which is birationally equivalent to V, and classifies all curves of degree 3 on W. He mentions that the surface W has a superabundance of rational curves lying upon it. F. Beukers, van Geemen, < 2000: In their (unpublished) preprint, they show that V is birationally equivalent to an elliptic surface V fibered over P 1. The sections of this fibration give rise to infinitely many rational curves on V and hence on V.
3. New Ideas (using Arithmetic Geometry) - 2 R. van Luijk, 2000: In his (unpublished) thesis, [vl] studies the box variety (the name is due to [FS]) B P 6 defined by equations (1) (4), whose rational points B(Q) (with abc 0) correspond to perfect cuboids. He proves that B C is a normal surface with 48 singularities, and that its desingularization B is of general type. He also finds a set L of 92 curves of genus 1 on B C : 24 curves/q isomorphic to P 1 (the components of abc = 0,) 8 curves/q(i), isomorphic to P 1, 60 elliptic curves/q( 2, i), none defined over Q. Thus: The geometry of the underlying surfaces V and B is radically different.
4. The Bombieri-Lang Conjecture Question: What special diophantine properties do varieties of general type have? Theorem of Faltings, 1983: If C/K is a curve of general type ( g C 2) over a number field K, then C(K) is finite. Bombieri-Lang Conjecture: If X /K is a variety of general type, then X (K) is not Zariski-dense in X. (K a number field.) Lang Conjecture (LC): If X /K is a variety of general type, then there is a proper closed subset E(X ) X such that U(X ) := X \ E(X ) has finitely many K -rational points for every number field K /K. Remark: If (LC) is true for a surface X /K, then we must have: E(X ) = union of all genus 0 and 1 curves on X K. Since E(X ) is supposed to be a closed set, this implies:
4. The Bombieri-Lang Conjecture - 2 Geometric Lang Conjecture (GLC): A surface X /K of general type contains at most finitely many curves of genus 1. Theorem of Bogomolov, 1977: (GLC) is true for a smooth surface X /C, provided that c 2 1 (X ) > c 2(X ). Difficulties: 1) How can we determine the exceptional set E(X )? Even in the situation of Bogomolov, there is no algorithm for determining E(X ). 2) The desingularization B C of the box variety does not satisfy Bogomolov s hypothesis. (Here c 2 1 ( B C ) = 16, c 2 ( B C ) = 80.)
5. Further Results - all unpublished, but available on arxiv [math.ag]. M. Stoll, D. Testa, 2010: study the box variety B and its desingularization B in detail. For example: they compute all the geometric invariants of B C = B C, they determine Aut( B C ) (so Aut( B C ) = 1536 = 2 9 3), they prove that the Picard group Pic( B C ) = NS( B C ) Z 64 and is generated by 140 curves: the 92 curves found by [vl], and the 48 exceptional curves which resolve the 48 singularities. Question/Conjecture[ST]: Is E(B C ) = L? Is every curve of genus 1 on B C one of the 92 curves found by [vl]? Note: Conjecture[ST] + Lang s Conjecture (LC) (+ [vl]) imply that Problem 3 has a positive answer: there are only finitely many perfect cuboids.
5. Further Results - 2 A. Beauville, 2013: gives a more intrinsic construction of the box variety B C. This implies: B C is a diagonal quotient surface! E. Freitag, R. Salvati Manni, 2013: give an analytic and a modular description of B C, Aut(B C ) and of L. 1) They construct B C as an (explicit) quotient of H H ; 2) They show that (an open part of) B Q(i) has a modular interpretation; 3) All α Aut(B C ) are modular (i.e., induced by Γ(1) Γ(1)); 4) All curves in L are modular (or cuspidal). In particular: B Q(i) is a (generalized) modular diagonal quotient surface.
6. Diagonal Quotient Surfaces Let: X be a smooth, projective curve over a field K, G Aut(X ) a finite group of automorphisms of X, π : X X := G\X, the quotient map, Y := X X, the product surface, G = {(g, g) : g G} G G, the diagonal subgroup, Z G := G \Y, the diagonal quotient surface defined by G, φ = φ G : Y Z G, the associated quotient map, ψ = ψ G : Z G Y := X X, the induced map. Remarks: 1) We have that ψ φ = π π : π π : Y = X X φ Z G ψ Y = X X. Thus, φ and ψ are finite of degree deg(φ) = deg(ψ) = G. 2) In general, Z G has finitely many quotient singularities. The structure of Z G and of its desingularization Z G can be worked out; cf. K.-Schanz, 1997.
6. Diagonal Quotient Surfaces - 2 Application: We apply this to: X := X (8) = Γ(8)\H, the modular curve of level 8, G = Γ(4)/Γ(8) (Z/2Z) 3, so X = X (4) P 1. Note: g X = 5, and X and G are defined over Q. Theorem 1: ( [FS]) Let Z G /Q be the diagonal quotient surface of X = X (8) /Q and G = Γ(4)/Γ(8). Then (5) B Q(i) Z G Q(i). Key Idea: [FS] use the fact that there is an isomorphism X (8) Proj(A), where A := C[ϑ 00 (z), ϑ 10 (z), ϑ 01 (z), ϑ 00 (2z), ϑ 10 (2z)]) is the graded ring generated by the classical theta-functions ϑ a,b (of weight 1 2 ).
6. Diagonal Quotient Surfaces - 3 Recall that the theta functions ϑ a,b are defined by ϑ a,b (z) = n= and satisfy the classical theta-relations e πi((n+a/2)2 z+b(n+a/2)). ϑ 2 00(z) = ϑ 2 00(2z) + ϑ 2 10(2z) ϑ 2 01(z) = ϑ 2 00(2z) ϑ 2 10(2z) ϑ 2 10(z) = 2ϑ 00 (2z)ϑ 10 (2z). [FS] also work out the action of the generators T = ( ) 1 4 0 1, T = ( ) ( 1 0 4 1, R = 5 8 ) 8 13 of G = Γ(4)/Γ(8) on the basis ϑ 00 (z), ϑ 10 (z),..., ϑ 10 (2z). This allows them them to compute the ring of invariants of A A with respect to G. (They use instead a group (4, 8), but this leads to the same ring of invariants.)
7. Modular Correspondences Question: What is special about a product of modular curves? Partial answer: It comes equipped with a rich supply of curves, the modular correspondences. ( Hecke operators.) Construction: Let α GL + 2 (Q) M 2(Z), and let T (α) = {(z, α(z)) : z H } H H be its graph. Then for any Γ = Γ(N), its image T Γ (α) X (N) X (N) = Γ(N)\H Γ(N)\H is an irreducible (algebraic) curve on X (N) X (N). Example: For Γ = Γ(1), each modular correspondence is of the form T Γ (α n ), for some n 1, where α n = ( 1 0 0 n). Moreover, T Γ (α n ) X 0 (n).
7. Modular Correspondences - 2 Theorem 2: ([FS]) Via the isomorphism (5), each curve in L is either the image of modular correspondence T (α) (with det(α) = 1) or the image of cuspidal curve {c} H or H {c}, where c H \ H is a cusp. The following is a partial converse: Theorem 3: If C is curve of genus 1 on B C Z G C which is modular (i.e., C = φ(t Γ(8) (α)), for some α), then C L. Thus: The Conjecture [ST] (i.e., E(B) = L) (*) every curve in E(B) is modular or cuspidal.
8. Mazur s Question Mazur, 1978: To what extent are the isogeny classes of elliptic curves E/Q determined by their mod N Galois representations? Frey (1985), Darmon (1994): Formulated conjectures which make this question much more precise. Remark: To study isomorphisms of mod N Galois representations, one is naturally led to consider the modular diagonal quotient surface Z N = Z X (N),GN, where G N = Γ(1)/(±Γ(N)), because the points of Z N classify such isomorphisms. Conjecture (1995): If N = p is a prime with p 23, then every curve C in E(Z p ) is modular. Note: This conjecture + (LC) Darmon s Conjecture.
8. Mazur s Question - 2 Theorem 4 (Bakker/Tsimerman, 2013): The above conjecture is true for p >> 0. Remark: Unfortunately, their proof does not give any estimate on how large p has to be. But suppose: that their methods could be refined to give a proof of the above conjecture. Then it might be possible to also prove a similar statement for Z G C = B C, i.e. to prove condition (*). Then we would have: (LC) #(perfect cuboids) < (Problem 3)